[NFC] Optimize ModuleUtils type-scanning code (#8572)

Before, we always did a loop on `type.getHeapTypeChildren()` which
means setting up a scanner object and going through a generic
path. Instead, handle the common cases directly. This avoids any
generic path in the common case.

This makes us 1% faster on `-O3`, as measured by instruction
count, number of branches, and walltime. The noise in the first
two is incredibly small, so this looks reliably faster.

Author: kripken

src/ir/module-utils.cpp

@@ -363,8 +363,15 @@ struct TypeInfos {
     }
   }
   void note(Type type) {
-    for (HeapType ht : type.getHeapTypeChildren()) {
-      note(ht);
+    // Handle the common case of a ref directly, to avoid a scan of children.
+    if (type.isRef()) {
+      note(type.getHeapType());
+      return;
+    }
+    if (type.isTuple()) {
+      for (HeapType ht : type.getHeapTypeChildren()) {
+        note(ht);
+      }
     }
   }
   // Ensure a type is included without increasing its count.
@@ -374,8 +381,14 @@ struct TypeInfos {
     }
   }
   void include(Type type) {
-    for (HeapType ht : type.getHeapTypeChildren()) {
-      include(ht);
+    if (type.isRef()) {
+      include(type.getHeapType());
+      return;
+    }
+    if (type.isTuple()) {
+      for (HeapType ht : type.getHeapTypeChildren()) {
+        include(ht);
+      }
     }
   }
   void noteControlFlow(Signature sig) {